Document the work-program participation inputs as unsourced assumptions

[daphnehanse11]

Closes #249; advances #248. Stacked on #352 (draft until it merges, then rebases to main — the diff vs #352 is one register extension plus contract tests).

Summary

The CPS ASEC has no item measuring SNAP E&T, qualifying work-program, or workfare participation — confirmed against both the raw ASEC column inventory and the retired enhanced-CPS pipeline, which never carried these inputs either. Per #249's scope ("if no source exists, add an explicit release diagnostic/documented assumption rather than relying on an invisible default"), this adds the three work-program inputs to the documented_absent_inputs register introduced in #352:

• is_snap_work_program_participant — general work-registration compliance via program participation never fires; USDA reports E&T reaches a small minority of participants, so the always-False default understates compliance only modestly.
• weekly_snap_work_program_hours — combined work-plus-program hours toward the 20-hour ABAWD test omit program hours.
• is_snap_workfare_participant — workfare under 7 CFR 273.7(m) satisfies the ABAWD requirement regardless of hours; defaults False.

Contract tests

New test_us_documented_absent_inputs.py pins the register: exact coverage of the six known unsourced inputs, every entry citing a tracking issue, disjointness from the degenerate-exclusions register (persisted-stuck vs never-persisted classes must not overlap), and no seeded input (pregnancy, disability, discretionary exemptions) listed as absent.

Why documentation rather than seeding

Unlike pregnancy (#352) or discretionary exemptions (#353), there is no statutory rate or stable administrative share to seed against at the person level — E&T participation is state-program-specific and small. If a citable assignment basis emerges (e.g., state E&T participation counts), the register entry converts to a seeded stage the same way; until then the honest treatment is a visible always-False assumption.

🤖 Generated with Claude Code

[daphnehanse11]

#352 merged, so this is rebased onto main and out of draft — the diff is now exactly the register extension (three work-program entries) plus the contract tests.

[MaxGhenis]

Approve

Verified each new register entry against the underlying policyengine-us variables (installed build pin, gov/usda/snap/eligibility/work_requirements/). All three are real, unseeded, and the reasons are accurate — two of them match the variable's own documentation on the load-bearing citations.

• weekly_snap_work_program_hours (float, defaults 0): variable doc states these hours "count toward the 20-hour weekly ABAWD work-activity threshold, alone or combined with hours of employment ... Survey data does not capture this input, so it defaults to zero; see #249." The register reason ("combined work-plus-program hours toward the 20-hour ABAWD test omit program hours") is exactly this. Feeds meets_snap_abawd_work_requirements.
• is_snap_workfare_participant (bool, defaults False): variable doc cites "7 CFR 273.7(m)" and "satisfies the SNAP ABAWD work requirement regardless of the number of hours." The register reason reproduces both facts verbatim. Feeds meets_snap_abawd_work_requirements.
• is_snap_work_program_participant (bool, defaults False): per its doc, this drives general work-requirement compliance (7 CFR 273.7(a)(1)) and explicitly "does not satisfy the ABAWD work requirement," so it flows to meets_snap_general_work_requirements. The "always-False default understates compliance only modestly" framing is directionally correct (defaulting False marks would-be E&T compliers as non-compliant) and appropriately hedged for an assumptions register; #249 tracks the sourcing.

Confirmed absent, not degenerate: none of the three columns is seeded or persisted anywhere in packages/ or tools/, so the disjointness with US_DEGENERATE_INPUT_REVIEWED_EXCLUSIONS is real, not just asserted. The comment edit generalizing "exemption channels" to "compliance/exemption channels" is correct — the #351 inputs drive exemption, the new #249 family drives compliance.

Pattern matches #352/#351: entries added to US_DOCUMENTED_ABSENT_INPUTS with issue-citing reasons, plus a contract test that pins the register to exactly the six known unsourced inputs, requires every reason to cite a tracking issue, and enforces disjointness from both the seeded stages and the degenerate-exclusions register. CI green (test + wheels, 3.13/3.14), MERGEABLE/CLEAN.

Not merging (review-only per request).